Secret Key Generation for Secure Wireless Internet of Things

In the Internet of Things (IoT), hundreds or even thousands of devices with wireless connectivity and typically equipped with sensors are communicating over shared radio spectrum. Applications of IoT include smart homes and cities, monitoring environment and radio spectrum, machine-type communications, and well-being. Ensuring the security of data in IoT is a serious challenge. Existing secrecy technologies, such as traditional encryption based on computational secrecy, are seen as too computationally demanding. In particular, IoT devices are often battery-powered and must be energy efficient, imposing constraints on computation, communication and price. As a potential alternative to traditional encryption, the concept of physical-layer security has been proposed. Since it is implemented on the physical layer of a wireless system, rather than the application layer, physical-layer security is in principle no more demanding than non-secret communication. Instead of computational secrecy, physical-layer security is based on informationtheoretic secrecy, which has provable secrecy guarantees regardless of an adversary's computational power. Thus, information-theoretic secrecy cannot be broken by e.g. quantum computers. On the other hand, it also has a number of drawbacks that may make its implementation challenging in practice, such as relying on knowledge of the adversary's channel. In this thesis, theory and methods for hybrid secrecy systems are developed by combining both information-theoretic secrecy and computational secrecy. Hence, many of the drawbacks of using either system in isolation may be avoided. In such a hybrid system, a secret key (SK) is first generated based on information-theoretic secrecy, with the SK subsequently used in a lightweight symmetric-key encryption algorithm to achieve secure communication. The thesis focuses on the secret key generation (SKG) problem, particularly when it is subject to constraints on computational resources and communication. Short blocklength processing used in low-latency and high reliability communications is of particular interest. The contributions of the thesis are two-fold. First, theoretical bounds for the SKG problem in the finite-blocklength regime are established. Specifically, upper and lower bounds on the rate at which SKs can be generated are derived, both with and without constraints on communication. These new bounds refine existing bounds to yield more accurate information at short blocklengths. Second, practical SKG protocols to be used for IoT are designed. These protocols exploit the random and reciprocal nature of wireless channels to derive the key from channel coefficient estimates. One protocol is based entirely on the quantization of the channel estimates, with no error correction. This allows for a very low computation and communication overhead, at the cost of a higher bit error rate (BER) in the generated keys. Another protocol uses error correction based on polar codes to achieve competitive key rates at short blocklengths for arbitrary BERs.